Open-Vocabulary SAM: Segment and Recognize Twenty-thousand Classes Interactively

TL;DR

This paper introduces Open-Vocabulary SAM, a unified model combining SAM and CLIP for interactive segmentation and recognition of around 22,000 classes, significantly outperforming baseline methods.

Contribution

The paper presents a novel framework integrating SAM and CLIP with knowledge transfer modules, enabling large-scale open-vocabulary segmentation and recognition.

Findings

Effective knowledge transfer between SAM and CLIP demonstrated.

Achieves recognition of approximately 22,000 classes.

Outperforms naive baseline combinations in experiments.

Abstract

The CLIP and Segment Anything Model (SAM) are remarkable vision foundation models (VFMs). SAM excels in segmentation tasks across diverse domains, whereas CLIP is renowned for its zero-shot recognition capabilities. This paper presents an in-depth exploration of integrating these two models into a unified framework. Specifically, we introduce the Open-Vocabulary SAM, a SAM-inspired model designed for simultaneous interactive segmentation and recognition, leveraging two unique knowledge transfer modules: SAM2CLIP and CLIP2SAM. The former adapts SAM's knowledge into the CLIP via distillation and learnable transformer adapters, while the latter transfers CLIP knowledge into SAM, enhancing its recognition capabilities. Extensive experiments on various datasets and detectors show the effectiveness of Open-Vocabulary SAM in both segmentation and recognition tasks, significantly outperforming the na\"{i}ve baselines of simply combining SAM and CLIP. Furthermore, aided with image classification data training, our method can segment and recognize approximately 22,000 classes.